Products

The Roc series: outdoor AI nodes.

Three sizes, one architecture. Each node combines GPU compute, liquid cooling, power distribution, networking and software in a weather-protected enclosure.

Entry · Pilot

Roc-S

A compact single-cabinet node for pilots, single sites and first deployments. Roughly the footprint of a large outdoor HVAC unit.

GPU	4 × professional-grade inference GPUs
CPU	2 × server-class x86 processors
Memory	Up to 1.5 TB system memory
Storage	High-capacity NVMe for models and vector data
Cooling	Direct-to-chip liquid loop with outdoor heat rejection
Enclosure	Outdoor-rated, leak detection with automatic shutoff
Network	Dual high-speed uplinks, fibre-ready, out-of-band management

Enquire about Roc-S

Standard · Production

Roc-M

Our standard production cabinet with higher GPU density, redundant cooling components and heat-recovery output for larger sites.

GPU	8 × GPUs, configurable
Cooling	Redundant pumps and heat exchangers
Heat reuse	Optional warm-water output for building or process use
Power	Power capping to match site transformer capacity
Monitoring	Flow, pressure, temperature, leak, door and smoke sensing
Availability	Roadmap · enquire for timing

Enquire about Roc-M

Cluster · Scale

Roc-X

A multi-cabinet cluster sharing a central cooling plant and power train. Designed for campuses, telecom sites and large industrial parks.

Scale	Multiple Roc cabinets under one orchestration layer
Cooling	Shared central cooling plant, free-cooling where climate allows
Redundancy	Configurable N+1 or 2N for power and cooling
Delivery	Pre-integrated, factory-tested modules
Availability	Custom projects · enquire for timing

Enquire about Roc-X

Software

RocOS: the stack that makes the hardware useful.

Every node ships with a pre-installed, tested software stack so you can call an inference API the day it is powered on.

L5

Thermal-aware schedulingJob queueing and throttling tied to cooling state, site power and ambient conditions.

L4

Fleet consoleOne dashboard for every node: health, utilisation, updates and usage reporting.

L3

Model packsCurated open models for language, vision, embeddings and re-ranking. Bring your own fine-tunes.

L2

Inference servicesHigh-throughput serving behind an OpenAI-compatible API, with batch and real-time modes.

L1

RuntimeLightweight Kubernetes, GPU drivers and local storage services.

L0

Device and cooling controlBMC integration, sensor collection and automatic safety actions.

Built in

Designed for the outdoors.

Leak protection

Sensors, drip trays and automatic isolation valves guard against coolant leaks.

Weather resilience

Sealed, insulated enclosures protect against dust, rain and temperature swings.

Remote operation

Out-of-band management, remote power control and secure update channels.

Not sure which node fits?

Tell us your workloads and site conditions. We will recommend a configuration.

Talk to an Engineer